[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Discussion on global variables in IDL

Martin Schultz <martin.schultz@dkrz.de> wrote in message 
> alt@iszf.irk.ru (Altyntsev Dmitriy) writes:
> > Hello, 
> > 
> > 	I'd like to discuss global variable management in IDL. I've read a
> > pair of NG threads on that theme, and would like to add some
> > consideration and share experience. Some time ago I've sent a letter
> > to RSI with some proposals, but since then two version have been
> > released and I don't see any movement in the direction of improving
> > global vars management. It seems that everyone is satisfied. Is it
> > really so?
>   Well, I don't think the situation is as bad as you paint it. 
> First of all  I recommend objects  
> Cheers,
> Martin

My first reaction on yours answers was that I had been misunderstood.
I even post a message about it, but GOOGLE did not put it for some
reason in this thread. Then I started to analyze what I need and what
OOP offers me. I have read a pair of OOP books, scanned one more time
this NG as far as possible, looked up Martin's IDL_Container and
others. And to my great surprise discover that I have reinvented the
OOP wheel. It is funny (or sadly?) that I am using now procedure
naming conventions in the style of OOP like OBJ_Method. And I am using
pointer to data of my object in a style I have described before. I was
completely not right to name this problem as GLOBAL variable problem.
Global is global. But it is a problem how to share some data between
several procedures. And it actually is a problem of OOP implementing
in IDL. And having looked up through this NG I have found that this is
a problem in IDL and many people are NOT satisfied of current IDL OOP
state. Some of them have written their own 'containers', others do not
use OOP, and some even threatened with stopping to use IDL.

The most interesting NG thread (IMHO) was "Top 10 wishes". (BTW, I
could not find the end and the result of this epic work.) I wish this
column were persistent and wish to describe my current wish to IDL

So, what is the most annoying thing with IDL OOP for me? The fact that
when I describe class I loose freedom of changing field type and size.
I can not have undefined field in class too. And I can not add field
during execution. Let us view very simple typical example.

IND = WHERE(A GT 0) ; the essence of IDL style

Suppose we want this IND as field of some class. Then I need (as David
tired to repeat) to use pointers and write like this:

function class1::INIT
	self.ind = ptr_new(/all)
	return, 1
pro class1::CLEANUP
	ptr_free, self.ind
pro class1::method1
	*self.ind = where(A GT 0)
pro p
	struct = { class1, ..., IND:ptr_new(), ... }
	Obj1 = obj_new( 'class1' )
	obj_destroy, Obj1

So, instead of one simple string I must initialize and destroy pointer
manually. If during development I suddenly wanted to have IND as field
of the class, I must add it in class definition, in class init, and in
class cleanup. Or use some slow container, because all of them are
written in IDL with a lot of operations.

I do not understand why IDL can not do all this stuff itself. It can
be done several ways. As for me, I would like to form class fields
during execution. It seems to me on the whole of IDL style.

pro class1::method1
	self.IND = where(A GT 0)
	self.IND = 'We can change type of the field'
	tmp = temporary(self.IND) ; or destroy it	
pro p
	struct = { class1, public ..., IND, ..., privite ... }
	Obj1 = obj_new( 'class1' )
	obj_destroy, Obj1

As soon as IDL meet self.IND it inits it at once. And destroy by
obj_destroy or reassigning like local variables.

I think that it would be fine to have an opportunity even not define
variable in class description but it involves complicated symbol
tables managing by IDL during compile time or slow during execution.
It may be difficult task but solvable.

And of course, I must have an opportunity to assign class method as
event handler in widget applications.

What would you say?

The other my "Top 10" very shortly 
1) Keystroke events and MDI interface (ENVI and others apps might be
much more convenient). I use widget_text trick now.
2) Handy and effective tools for viewing images and plots (colors,
multichannel, cursor position, zoom, image enhancement, export,
printing, ...). I have my own tools. They are far from perfection.
3) Class explorer, so I could quickly look up my fields and methods.
Now I use external text editor search capabilities.
4) GIS functions to work with vector data (polygon intersections,
area, distance ...) Writing myself.
5) FTP object
6) Low or null prize of run-time package

This will be enough. 

Altyntsev Dmitriy